| # Layer 4 — Incremental Parser (15000 lines target) | |
| ## What This Layer Does | |
| Tree-sitter / GLR incremental parser with error recovery. Parses the rope from | |
| Layer 3 into a concrete syntax tree (CST). Re-parses only changed regions on | |
| each keystroke — O(edit size * grammar size), not O(file size). | |
| ## Repos That Contribute | |
| | Repo | What to pull | | |
| |------|-------------| | |
| | **snapkitty-resonance-isa** | Abjad VM IR bytecode — maps directly to parser stack machine | | |
| | **snapkitty-resonance-bridge** | proof→SUBLEQ encoding — extract as parser IR substrate (EXTRACT) | | |
| | **j-matrix-twin** | SUBLEQ attention + WASM (EXTRACT) — alternative parser substrate for browser | | |
| | **systemic-intelligence** | GLR skeleton in OCaml section | | |
| ## What to Build First | |
| 1. Grammar representation — rule table, terminal/nonterminal IDs (use Layer 0 intern) | |
| 2. GLR parse table — state machine over the grammar | |
| 3. Incremental reparse — diff the rope edit, re-enter parser at affected range | |
| 4. Error recovery — skip/insert tokens, maintain partial tree | |
| 5. SUBLEQ substrate port — from j-matrix-twin for WASM browser path | |
| 6. Abjad VM IR — from snapkitty-resonance-isa as the hot-path parser stack | |
| ## Key Design Constraints | |
| - CST nodes in Layer 0 arena; GC via arena reset between full reparses | |
| - Edit regions from Layer 3 rope trigger incremental reparse only | |
| - Grammar tables are WORM-sealed after load (bifrost pattern from Layer 0) | |
| - Parser must run in O(n log n) worst case, not O(n^2) | |